[dpdk-dev] Service lcores and Application lcores

[Thomas Monjalon]

30/06/2017 15:24, Van Haaren, Harry:
> From: Jerin Jacob [mailto:jerin.jacob at caviumnetworks.com]
> > From: "Van Haaren, Harry" <harry.van.haaren at intel.com>
> > > From: Jerin Jacob [mailto:jerin.jacob at caviumnetworks.com]
> > > > From: "Van Haaren, Harry" <harry.van.haaren at intel.com>
> > > <snip previous non-related items>
> > >
> > > > > I don't think providing a remote-launch API is actually beneficial. Remote-launching
> > a
> > > > single service
> > > > > is equivalent to adding that lcore as a service-core, and mapping it to just that
> > single
> > > > service.
> > > > > The advantage of adding it as a service core, is future-proofing for if more
> > services
> > > > need to be added
> > > > > to that core in future, and statistics of the service core infrastructure. A
> > convenience
> > > > API could be
> > > > > provided to perform the core_add(), service_start(), enable_on_service() and
> > > > core_start() APIs in one.
> > > > >
> > > > > Also, the remote_launch API doesn't solve the original problem - what if an
> > application
> > > > lcore wishes
> > > > > to run one iteration of a service "manually". The remote_launch style API does not
> > solve
> > > > this problem.
> > > >
> > > > Agree with problem statement. But, remote_launch() operates on lcores not on
> > > > not necessary on 1:1 mapped physical cores.
> > > >
> > > > By introducing "rte_service_iterate", We are creating a parallel infrastructure to
> > > > run the service on non DPDK service lcores aka normal lcores.
> > > > Is this really required? Is there  any real advantage for
> > > > application not use builtin service lcore infrastructure, rather than iterating over
> > > > "rte_service_iterate" and run on normal lcores. If we really want to mux
> > > > a physical core to N lcore, EAL already provides that in the form of threads.
> > > >
> > > > I think, providing too many parallel options for the same use case may be
> > > > a overkill.
> > > >
> > > > Just my 2c.
> > >
> > >
> > > The use-case that the rte_service_iterate() caters for is one where the application
> > > wishes to run a service on an "ordinary app lcore", together with an application
> > workload.
> > >
> > > For example, the eventdev-scheduler and one worker can be run on the same lcore. If the
> > schedule() running thread *must* be a service lcore, we would not be able to also use that
> > lcore as an application worker core.
> > >
> > > That was my motivation for adding this API, I do agree with you above; it is a second
> > "parallel" method to run a service. I think there's enough value in enabling the use-case
> > as per example above to add it.
> > >
> > >
> > > Do you see enough value in the use-case above to add the API?
> > 
> > The above use case can be realized like --lcores='(0-1)@1'(Two lcore on
> > an physical core). I believe, application writers never want to write a
> > code based on specific number of cores available in the system. If they
> > do then they will be stuck on running on another environment and too
> > many combination to address.
> 
> Good point.
> 
> > For me it complicates service lcore usage. But someone think, it will useful then
> > I don't have strong objection.
> 
> We can easily add APIs later - and removing them isn't so easy. +1 from me leave it out for now, and we can see about adding it for 17.11 if the need arises.
> 
> Thanks for your input, I'll spin a v3 without the rte_service_iterate() function, and that should be it then!

I agree to leave it and keep things simple.